Global Growth Opportunities and Market Integration

Transcription

1 Global Growth Opportunities and Market Integration Geert Bekaert Columbia University, New York, NY USA National Bureau of Economic Research, Cambridge, MA USA Campbell R. Harvey Duke University, Durham, NC USA National Bureau of Economic Research, Cambridge, MA USA Christian Lundblad Indiana University, Bloomington, IN USA Stephan Siegel Columbia University, New York, NY USA University of Washington, Seattle, WA USA June 2005 We appreciate the helpful comments of Rajesh Chakrabarti, Will Goetzmann, John Graham, Anna Pavlova, Lijian Sun, Jeff Wurgler, and participants at the 2004 EFA meetings, the October 2004 Financial Market Integration lecture series at the ECB, the 10th Annual Global Investment Conference in Whistler, the 11th Assurant / Georgia Tech Conference on International Finance, the 2005 WFA meetings, the Globalization and Financial Services JBF/World Bank Conference, and the China International Conference in Finance. We are especially grateful for the extensive comments of the referee. Send correspondence to: Campbell R. Harvey, Fuqua School of Business, Duke University, Durham, NC Phone: , cam.harvey@duke.edu.

2 Global Growth Opportunities and Market Integration Abstract We propose an exogenous measure of a country s growth opportunities by interacting the country s local industry mix with global price to earnings (P E) ratios. First, we find that these exogenous growth opportunities strongly predict future changes in real GDP and investment in a large panel of countries. This relation is strongest in countries that have liberalized their capital accounts, equity markets, and banking systems. Second, we re-examine the link between financial development, external finance dependence, investor protection, capital allocation, and growth. We find that financial development, external finance dependence, and investor protection measures are much less important in aligning growth opportunities with growth than is capital market openness. Third, we formulate new tests of market integration and segmentation. Under integration, the difference between a country s local P E ratio and its global counterpart should not predict relative growth, but the difference between its exogenous global P E ratio and the world market P E ratio should predict relative growth. KEYWORDS: Growth Opportunities, Market Integration, Finance and Growth, Financing Constraints, Capital Allocation, Capital Account Openness, Financial Liberalization, Industry Mix, GDP Growth, Investment Growth.

3 1 Introduction In a perfectly integrated world economy, capital should be invested where it expects to earn the highest risk-adjusted return. Much of the research on real variables and quantities is strongly at odds with the notion of global integration. For example, in their classic study of 16 developed countries, Feldstein and Horioka (1980) found that domestic saving rates explain over 90% of the variation in investment rates. Because the Feldstein and Horioka sample ends in 1974, it does not reflect the considerable progress towards globalization in the 1970s and 1980s. However, Obstfeld and Rogoff (2000) continue to find a high correlation between domestic investment and savings for the period, both for the OECD countries and a group of mid-income emerging countries. Apart from a home bias in real investments, research has documented a home bias in trade. Even controlling for tariffs, a country is much more likely to trade within its own borders than with neighboring countries. 1 There is also a well-documented home asset bias. Despite uncontroversial diversification benefits, there is a strong preference for investing in domestic securities. 2 While the case for imperfect integration is strong when using real/quantity variables, it is more mixed when using prices and returns. For example, Harvey (1991) finds evidence that a global version of the CAPM cannot be rejected in almost all developed country equity markets (with Japan as the exception). For emerging markets, Bekaert and Harvey (1995, 2000) provide sharper evidence against the hypothesis of global equity market integration. The benefits of increasing globalization are now being questioned even though its welfare benefits may be large (see Lewis (1999) for the latter and Rodrik (1998) and Stiglitz (2000) for the former). We add a new perspective to the literature. Our research proposes a simple measure of country-specific growth opportunities based on two rather non-controversial assumptions. First, the growth potential of a country is largely reflected in the growth potential of its mix of industries. Second, price earnings (P E) ratios contain information about growth opportunities. If markets are globally integrated, we can measure a country s growth 1 See, for example, McCallum (1995) and Helliwell (1998). 2 See, for example, French and Poterba (1991), Tesar and Werner (1995), Baxter and Jermann (1997), and Lewis (1999). 1

4 opportunities by using the price earnings ratios of global industry portfolios weighted by the country s industrial mix. This perspective potentially offers a number of useful economic insights. First, for each country in the world, it permits the construction of an exogenous growth opportunities measure that does not use local price information. Such a measure should prove useful in numerous empirical studies seeking to avoid endogeneity problems. One example is the study by Bekaert, Harvey and Lundblad (2005), which examines the effect of equity market liberalization on growth. If countries liberalize when growth opportunities are abundant, regressions of future growth on a liberalization indicator suffer from a severe endogeneity problem. Measures of growth opportunities that use local price information are problematic because they may either reflect exogenous growth opportunities or better growth prospects induced by the liberalization decision. For the exogenous growth opportunities measure to be useful, it must actually predict growth. That is, countries that happen to have a high concentration of high P E industries should grow faster than average. We find that they do. Second, our framework can be employed to shed new light on the link between financial development, capital allocation, and growth (see Levine (2004) for a survey). Research by Rajan and Zingales (1998), Wurgler (2000), and LaPorta et al. (2000) stress the role of financial development in relaxing external financial constraints and improved investor protection as the critical growth channels. However, recent work by Fisman and Love (2004a, 2004b) suggests that financial development simply better aligns industry growth opportunities with actual growth. We test this hypothesis directly in a panel framework, in contrast to the purely cross-sectional approach followed in the existing literature. Moreover, the literature implicitly ignores the role of international capital flows. We also investigate how important financial openness is for aligning growth opportunities with growth. If financial openness is effective, countries that have liberalized their capital accounts, equity markets, or banking sectors, should display a closer association between growth opportunities and future real activity. Third, our measure can be used in formal tests of market integration that bridge research 2

5 on real quantities with price-based variables. When growth opportunities are competitively priced and exploited in internationally integrated markets, industry P E s should be equalized (barring risk differences) across countries. Consequently, under the null of market integration, the difference between a country s industry weighted global P E ratio and the world market P E ratio should predict future real GDP growth relative to world growth. Conversely, the difference between a country s global and local PE ratio should not predict growth in excess of world growth. We also investigate how these integration tests depend on measured degrees of financial openness. The remainder of the paper is organized as follows. Section two motivates our growth opportunities measure using a simple present value model, details its construction, its link with market integration, and provides some summary statistics. The third section investigates whether our growth opportunities measures indeed predict GDP and investment growth, contrasting the predictive performance of local with global measures. In the fourth section, we compare the different roles of financial openness, financial development, external finance dependence, investor protection, and political risk in aligning growth opportunities with growth. The fifth section formulates and conducts our test of market integration. Some concluding remarks are offered in the final section. 2 Measuring Growth Opportunities 2.1 Growth opportunities, market integration, and economic growth Holding a number of factors such as risk constant, higher price earnings ratios indicate high growth opportunities. Others have proposed different proxies for growth opportunities. The corporate finance literature often uses market to book value as a proxy for Tobin s Q and a measure of investment opportunities (see e.g. Smith and Watts (1992), Booth et al. (2001), and Allayannis et al. (2003)). Fisman and Love (2003) and Gupta and Yuan (2004) use historical sales growth of U.S. industries as a measure of growth opportunities. In contrast to sales growth, P E has the advantage of being forward looking. In integrated world capital markets, all available growth opportunities are competitively 3

6 priced and exploited. This implies that a country s P E ratio for a particular industry should not significantly differ from its world counterpart. However, there are obviously different growth opportunities across industries. Hence, one source of local GDP growth relative to world GDP growth is the weighting of industries within a particular country. If a country has a large concentration in high P E (high growth opportunity) industries, then it should grow faster than the world. To formalize this, we view each country as consisting of a set of industries that receive stochastic growth opportunities. Under full integration, all opportunities are competitively priced and exploited, so there is no priced country-specific growth opportunity for any industry. Let (logarithmic) earnings growth be denoted by ln(ea t ) and let countries and industries be indexed by i and j, respectively. Assume ln(ea i,j,t ) = GO w,j,t 1 + ɛ i,j,t, (1) where GO w,j,t 1 represents the stochastic growth opportunity for each industry j which does not depend on the country to which the industry belongs. In contrast, ɛ i,j,t is a country and industry specific earnings growth disturbance. The assumption in (1) is strong and goes beyond financial market integration. Essentially, we assume economic integration to imply that industry earnings growth processes share a common component across countries. It is conceivable that non-tradable and regulated sectors in financially and even reasonably economically integrated countries still face priced country-specific growth opportunities. We investigate this possibility in Section 3.3. Conversely, it is conceivable that even when financial markets are closed, foreign direct investment (FDI) flows induce common components in earnings growth across countries. The discount rate process for each industry (δ i,j,t ) is an affine function of the world discount rate (δ w,t ), as would be true in a financially integrated market: δ i,j,t = r f (1 β i,j ) + β i,j δ w,t, (2) where β i,j represented the exposure to systematic risk for industry j in country i. In addition, suppose that β i,j = β j. (3) 4

7 That is, industry systematic risk is the same across integrated countries. Of course, this assumption will not hold if there are leverage differences across countries. For quite general dynamics for δ w and GO w,j, but with normally distributed shocks, Appendix A derives (in closed-form) the P E ratio as an infinite sum of exponentiated affine functions of the current realizations of the growth opportunity (with a positive sign) and the discount rate (with a negative sign). While the resulting expression is unwieldy, it can be linearized to yield: pe i,j,t = ā i,j + b i,j δ w,t + c j GO w,j,t, (4) where pe is the log P E ratio. Under full integration, b i,j = b j, and c j does not depend on country i because of the assumption in (1). Why do certain countries grow faster than the average? In a fully integrated world, there are only two channels of growth for a particular country: luck (the error term) and an industry composition that differs from the world s. These assumptions also imply that industry P E ratios are similar across countries as they are determined primarily by global factors. 3 Global industry P E ratios therefore contain the same information about industry growth opportunities in a given country as local P E ratios. As a consequence, as local and global industry P E ratios move together, the difference between them should contain no information about the country s future economic performance relative to the world economy. In contrast, this is not true when markets are not fully integrated and growth opportunities are not priced globally (but locally). That is, the link between our growth opportunities measures and future growth can lead to a test of market integration. Let P E i denote the vector of industry P E ratios in country i and P E w the vector of world industry P E ratios. Similarly, define country and world industry weights by IW i and IW w, respectively. Combining these vectors for country i, we define local growth opportunities 3 There is a country-specific intercept that comes from volatility terms and a potentially country-specific component to the discount rate, but the time variation in the P E ratio is driven by global factors. However, if there are systematic leverage differences across countries, P E ratios across countries will react differently to changes in global discount rates. 5

8 (LGO) and global growth opportunities (GGO): LGO i,t = ln[iw i,t 1P E i,t ] (5) GGO i,t = ln[iw i,t 1P E w,t ]. (6) In integrated markets, LGO and GGO reflect the same information and should hence both predict economic growth in country i. Furthermore, the difference between the two measures, which we call local excess growth opportunities (LEGO), should be constant and should therefore have no predictive power for relative economic growth. If markets are not fully integrated, though, LGO and GGO will display different temporal behavior and LEGO should predict economic growth in country i in excess of world economic growth. In other words, under our auxiliary assumptions, the hypothesis of no predictability constitutes a market integration hypothesis. If, on the other hand, we start from the hypothesis that markets are completely segmented, we do not expect global industry P E ratios to contain information about local growth opportunities. Hence, GGO should not necessarily predict economic growth in country i. Moreover, let s define the difference between GGO and the log of the world market price earnings ratio (W GO) as: GEGO i,t = GGO i,t W GO t (7) where W GO t = ln[iw w,t 1P E w,t ]. (8) Under the null of market segmentation, GEGO should not predict relative growth in country i as global prices contain no information about exploitable growth opportunities. If the hypothesis of market segmentation is incorrect, GEGO should predict economic growth in country i relative to world economic growth, because it reflects the difference between local and global industry composition. Under the above assumptions of market integration, this difference should be the only measure predicting relative growth. Predictive regressions of future relative economic growth onto GEGO allow us therefore to also test the hypothesis of market segmentation. Table 1 summarizes the proposed measures of growth opportunities as well as their ability to predict economic growth under different assumptions. 6

9 2.2 Constructing the growth opportunities measures We construct the measures of growth opportunities discussed above for a sample of 50 countries, listed in Appendix Table A1. To construct LGO and W GO, we simply take natural logs of the country-specific or world market P E ratio. We use monthly P E ratios from Datastream as the primary source. A few countries in our sample are not covered by Datastream and we use P E ratios from S&P s Emerging Markets Data Base (IFC) instead. For Italy, Norway, Spain, and Sweden, we use P E ratios from MSCI to exploit the longer time series compared to Datastream. For the construction of our exogenous measure of growth opportunities, GGO, we require global industry P E ratios as well as country-specific industry weights. We obtain global industry P E ratios for 35 industrial sectors with 101 sub-sectors from Datastream. construct two alternative sets of country-specific industry weights: the first uses equity market capitalization and the second uses a measure of value added to construct relative weights. Most of the results in the paper are based on the market capitalization weights. For 21 of our 50 countries, our measure simply uses the Datastream data to calculate the market capitalization of a country s industries relative to the country s total stock market capitalization for 35 industries. For the remaining 29 countries, we use the 82 SIC industry groups used by S&P s Emerging Markets Data Base (IFC) to determine a vector of industry weights. The local weights for these SIC industry groups are matched with the Datastream price earning ratios by linking the 101 Datastream sub-sectors to the corresponding local market industry structure. 4 As a robustness check, we also present results based on the alternative value-added weighting. 5 We use value added data from the UNIDO Industrial Statistics Database which covers 28 manufacturing industries in a large number of countries. The weight of an industry in a given country is determined by the industry-specific value added relative to the 4 An alternative way to merge the two industry classifications is to link the SIC industry structure used by Standard & Poor s to the 35 Datastream industries sectors to create a uniform vector of weights across all countries in our sample. This alternative method yields very similar growth opportunities measures. 5 A full set of results based upon the value-added weighting is available upon request. We 7

10 total value added of the manufacturing sector in that country. We again match the Datastream price earnings ratios to the 28 manufacturing industries used by UNIDO. 6 Appendix B provides much more detail about the construction of all measures of growth opportunities. Our tests may have low power when discount rate changes dominate the variation of the P E ratios. Therefore, we create an alternative measure by removing a 60-month moving average from the standard measure. For example, we define LGO MA as: LGO MA i,t = LGO i,t 1 60 t 1 s=t 60 LGO i,s (9) The relative measure is less likely to be dominated by discount rate changes, if discount rates are more persistent than growth opportunities, for which there is some empirical evidence. GGO MA, LEGO MA, and GEGO MA are calculated analogously. While some of our growth opportunities measures are available at a monthly frequency from as early as January 1973 until December 2002, the starting points for measures using local P E ratios vary across the 50 countries and other macro variables are available only at an annual frequency. Therefore, we only use the December values of our growth opportunities measures from 1980 until In addition to the complete set of the 50 countries, we study the subset of 17 developed countries for which we are able to construct LGO and LEGO for all years between 1980 and We also consider a subset of 30 emerging market countries for which the LGO and LEGO time series are of varied length. Table 2 provides a summary of the construction of all the variables and the data sources. 2.3 Comparing the growth opportunities measures Table 3 contains summary statistics for our growth opportunities measures. Panel A presents summary statistics for our unadjusted growth opportunities measures, averaged over different country groups and per country. The measure of local growth opportunities, LGO, is based on local P E ratios. Not surprisingly, it exhibits substantial time-series variation. It 6 In a related application, Almeida and Wolfenzon (2004) use the UNIDO weights and world industry measures of external financing needs, to construct an exogenous measure of a country s external financing needs. 8

11 exhibits substantial cross-sectional variation as well, being lower than 2 for Zimbabwe, Jamaica, Israel, and Côte d Ivoire, but higher than 3 for Italy and Japan. GGO, our measure of exogenous growth opportunities, shows lower dispersion than LGO. When comparing the sample of developed countries to the emerging market sample, we find little difference in the means and standard deviations of LGO and GGO. LEGO, the industry-weighted difference between information contained in local and global P E ratios, is on average higher in developed countries ( 0.208) than in emerging market countries ( 0.494). Similarly, GEGO has a higher mean in the sample of developed countries (0.044 vs ), possibly reflecting a more favorable industrial composition in developed countries. The variability of LEGO and GEGO is lower in the sample of developed countries than in the sample of emerging market countries, where countries such as Kenya and Israel have very high standard deviations. We report the same statistics for the exogenous growth opportunities measure based on the value-added weights (GGO(V A)) producing similar findings. Panel B reports the same summary statistics for the adjusted growth measures, that is the original measures less a 60-month moving average. With the exception of LGO M A, the same pattern as in Panel A emerges. LGO MA appears to be lower and more volatile in emerging market countries compared to developed countries. Remember, though, that the availability of local P E ratios is limited for emerging countries, so that the summary statistics for measures of local growth opportunities are not directly comparable across the two samples. Panel C presents correlations between the different unadjusted as well as adjusted measures of growth opportunities. In both cases, the correlations between LGO and W GO and LGO and GGO are substantially higher for developed countries than for emerging market countries. For several emerging markets, including Brazil, Israel, and Venezuela, the correlations are negative. The correlation between GGO and W GO is high for all countries, confirming that changes in GGO are mainly driven by changes in the global P E ratios rather than by slowly evolving industry weights. The final column reports in each case the correlation between our market capitalization based measure of exogenous growth opportunities and the alternative measure that uses value-added weights. In case of the unadjusted growth 9

12 opportunities measure, the correlation is, on average, 0.79 and never falls below Tunisia has the lowest correlation of Finally, Panel D reports some characteristics of local stock markets and the main industries in each market. Our sample includes well established stock markets in both the developed (US, UK, Switzerland) and developing markets (South Africa, Malaysia) and vice versa. Because the level of stock market development may affect the representativeness of our industry weights for the whole economy, the robustness check using the value-added weights becomes even more important. The top three industries represent typically more than 50% of total market capitalization and in over 35% of the countries the banking sector is the top industry. The second most prominent industry is oil and gas, finishing first in 15% of the cases. To investigate a potential trend towards increased international integration over the past 20 years, Figure 1 shows the evolution of the average absolute value of LEGO, i.e. the distance between LGO and GGO for the sample of developed countries. While noisy, there appears to be a downward trend in the annual sample average, consistent with increasing market integration. Still using only observations from developed countries, we run a regression of the absolute value of LEGO onto a (country-specific) constant and a time trend. We find a negative ( ) and highly significant trend coefficient (standard error: ), confirming a reduction in the distance between LGO and GGO for our sample of developed countries. We expect local and global measures of growth opportunities to converge when countries become more integrated, but we have no such prior with respect to GEGO (the difference between GGO and the log of the world price to earnings ratio (W GO)). Figure 2 shows that for developed as well as emerging market countries the absolute value of GEGO seems to have decreased over time up until about One potential explanation is that differences in industrial composition across countries have decreased over time. To explore this further, we measure the difference between a country s industrial composition and the world s industrial composition. For each developed country, we calculate the average absolute value of the differences between the country s 10

13 industry weights and the world s industry weights for each year. Figure 3 shows that differences between local and world industrial composition have decreased over time. 7 For some countries this process is more pronounced. For example, the industrial composition of the Austrian economy has moved substantially closer to the world s industrial composition. On the other hand, the relative industrial composition of the U.S. has changed little. Given its economic weight in the world economy, this is, of course, not surprising. Importantly, the figure shows that on average a country s industrial composition differs substantially from the world s industrial composition. Under the null of market integration, cross-sectional variation in this composition is the only factor that explains cross country growth differences. 3 Do Growth Opportunities Predict Growth? 3.1 Econometric framework The first regressions we consider are y i,t+k,k = α i,0 + α i,1,t LGO MA i,t + η i,t+k,k (10) y i,t+k,k = α i,0 + α i,1,t GGO MA i,t + η i,t+k,k (11) where y i,t+k,k is the k-year average growth rate of either real per capita gross domestic product or investment for country i. We run similar experiments using LGO i,t and GGO i,t as the regressors. 8 We follow the convention in the growth literature employing k = 5 to minimize the influence of higher frequency business cycles in our sample. We maximize the time-series content of our estimates by using overlapping five-year periods. We include country specific fixed effects, α i,0, consistent with the model in Section 2, to capture cross-sectional heterogeneity and potentially omitted variables. Regressions (10) 7 See Carrieri, Errunza, and Sarikissian (2004) for a similar result. 8 We also consider a risk-adjusted growth opportunities measure. We regress each global industry P E ratio onto the conditional world market variance, estimated as a GARCH(1,1) model, and then take the intercept and residual as the risk-adjusted P E ratio. Combining these adjusted global industry P E ratios with the corresponding industry weights, we obtain a risk-adjusted growth opportunities measure for each country. The evidence (not reported) is qualitatively unchanged. 11

14 and (11) both test whether our growth opportunities measures indeed predict growth. In Sections 3.2 and 3.3 respectively, we discuss estimates of α i,1,t forcing it to be an identical time-invariant constant across all countries. However, the GGO-measure should only predict growth in integrated markets. Therefore, in Section 3.4 we model the slope coefficient α i,1,t as a linear function of various measures of openness, with the parameters constrained to be identical in the cross-section. We employ the pooled time-series, cross-sectional (panel) Generalized Method of Moments estimator presented in Bekaert, Harvey, and Lundblad (2001). Standard errors are constructed to account for cross-sectional heteroskedasticity and the overlapping nature of the growth shocks, η i,t+k,k. This estimator looks like an instrumental variable estimator but it reduces to pooled OLS under simplifying assumptions on the weighting matrix. 3.2 Local growth opportunities Table 4 (Panel A) presents estimates for α i,1 in regression (10) for each of our three samples (fixed effects are not reported) for both GDP and investment growth. We use both LGO and LGO MA. Unfortunately, the time-series history on local market P E ratios is limited (see Appendix Table A1); hence, we report estimates for an unbalanced panel, maximizing the sample history for each country. Overall, country-specific growth opportunities, as measured by local P E ratios, are informative about future economic activity. For example, the estimates for all countries suggest that, on average, a one standard deviation increase in local growth opportunities, that is an increase of in LGO MA, is associated with a 17 and 60 basis point increase in annual output and investment growth, respectively. The estimated effect is somewhat more pronounced for the developed markets than the general case (all countries), but in both cases highly statistically significant. For the emerging markets, the association is positive, but weak economically and not uniformly significant. There are many possible reasons for this apart from a true lack of predictive information. First, our sample histories are more limited for emerging markets. Second, our tests may have less power for emerging markets because other factors, such 12

15 as political risk or structural changes (market reforms for instance) may be relatively more important in driving P E ratios than growth opportunities. Finally, the stock market in these countries is generally smaller and less representative of the total economy compared to developed markets. To explore further the idea that country-specific stock market characteristics may affect the predictive impact of local P E ratios, in Table 4 (Panel B) we repeat the regression, interacting the LGO measures with several country-specific variables. For example, certain markets may have more regulated sectors, making the market s P E ratio less useful for these countries. When we interact the LGO measure with the proportion of the market capitalization accounted for by industries less likely subject regulation (see Appendix Table A3 for details), we find a positive and significant interaction effect for the LGO measure but not for the LGO MA measure. We also interact the LGO measure with equity market turnover, an indicator of the liquidity and perhaps efficiency of the local stock market, but do not find the expected positive interaction effect. Finally, the local P E ratios may represent a cross-sectional heterogeneous and time-varying mix of local and global prices because of the presence of ADRs. For example, ADRs have been more prevalent in Latin-America than in South-East Asia and ADRs were of much less importance earlier in the sample. Given that local prices partially reflect a corporate governance, segmentation and illiquidity discount, while ADR prices do not, the total P E ratio may be not very informative about growth opportunities. We use the Levine and Schmukler (2003) measure of the degree of internationalization of different stock markets measured as the market capitalization of firms that cross list, issue ADRs or GDRs, or raise capital in international markets relative to total equity market capitalization. 9 Unfortunately, these data are only available from When we interact the LGO measures with the ADR measure, the constant term is positive and significant but the interaction term is negative, albeit not always statistically significant. When we extend the data on internationalization to the full sample using country-specific information in the trend towards internationalization, we find similar results (not reported). We conclude that local P E ratios house information about future growth opportunities, 9 We thank Sergio Schmukler for making these data available to us. 13

16 but their information content is limited for emerging markets, partially due to limitations in the data set, partially because local P E ratios are confounded by country-specific factors. 3.3 Global (exogenous) growth opportunities In Table 5 (first two lines), we test whether exogenous growth opportunities predict real GDP and investment growth. Recall that GGO and GGO M A reflect the industrial composition within each country in accordance with the growth opportunities available to those industries in the global market. In this case, we obtain estimates for a full balanced panel across all three samples. Overall, the global growth opportunities measure appears to be a strong, robust, and significant predictor of future output and investment growth in all samples. For example, the estimates for all countries suggest that, on average, a one standard deviation increase in global growth opportunities, that is an increase of in GGO MA, is associated with a 28 and 78 basis point increase in annual output and investment growth, respectively. For the developed markets, the predictive power of the global measure is slightly weaker than the local measure (see Table 4) for the level measures but stronger for the measures with a past moving average removed. For emerging markets, the predictive power of the global measure is significantly better than the local measure, especially for investment growth, with the coefficients always statistically significantly different from zero. Consequently, even though emerging markets may be segmented from global capital markets, local P E ratios in emerging markets do a poorer job predicting future growth opportunities than global P E ratios. Panel A contains two pieces of additional information. First, we consider the impact of an alternative industry weighting scheme. Second, we consider a third grouping of countries - the European Union. Local market capitalization data may not be fully representative of a country s real activity, for instance, they may be biased towards industries more likely to chose equity financing in bank-oriented economies. Therefore, for manufacturing industries we create industry weights using the value added information from the UNIDO Industrial Statistics Database. For the developed markets, this strengthens the predictive power of the level 14

17 measures, but weakens the predictive power of the MA-measures. The growth opportunities measures continue to strongly predict future growth. For emerging markets, where perhaps we would have expected the stock market based weights to be least informative, the valueadded measures actually show somewhat less but still overall strong predictive power for future growth. In future tables, we focus on the market capitalization based measures of exogenous growth opportunities. The evidence for the value-added measures is similar and is available upon request. We focus on a subset of countries from the European Union (plus Norway and Switzerland), which represent a relatively well-integrated set of countries where global growth opportunities should be particularly relevant for future growth. We find that the coefficients for the EU countries are very similar to what we find for developed countries. In Panel B, we explore whether the predictability depends on three local factors. We only do this for the All Countries sample. First, we exclude regulated industries in the construction of GGO. Appendix Table A3 lists those industries we view as likely regulated. Regulated industries are presumably less capable of exploiting global growth opportunities. We indeed find that the predictability is stronger when attention is restricted to unregulated industries, but the change in coefficients is rather minor. Second, we look at a subset of tradable industries. Appendix Table A3 again lists those industries we view as potentially non tradable. We expect tradable sectors to have a stronger link to the global economy and our growth opportunities measures to work better for this set of industries. Panel B reveals that, while the predictive power remains very strong, it is not overall stronger than for the full set of industries. Finally, many countries went through a process of privatization of state owned enterprises (SOEs), see Megginson and Netter (2001) for details. Given that state-owned companies are typically in industries such as mining that depend on global commodity prices and because they may represent a large part of the real economy, the degree of privatization that took place may affect the predictive power of the global growth opportunities measures. Rather than using privatization activity directly, we use the percent of economic activity accounted for by state-owned enterprises. Consequently, this variable is negatively correlated with 15

18 the degree of privatizations and is available in panel for 34 countries. When we interact the growth opportunities measure with this variable, we find highly significantly positive constants, and negative interaction coefficients as expected. However, the interaction coefficients are not statistically significantly different from zero. When we use an alternative SOE measure which reflects the proportion of workforce employed by SOEs (not reported), we do find significant interaction effects, but this measure is only available for 17 countries. One last experiment we conduct is to verify that the predictive power of our measure remains significant when we include year dummies or the log of the world market P E ratio (W GO). We find that both measures (equity marked and value-added based) are still informative about a country s future growth, discounting the possibility that their predictive power reflects a world wide wealth effect. 3.4 The effects of financial sector openness Many of the countries in our sample have undergone regulatory reforms that may have implications for the ability of industries to capitalize on the growth opportunities available to them. In particular, we focus on the liberalization of the capital account, equity market, and banking sector. Countries which are closed to foreign investors typically also restrict the ability of their firms to raise capital abroad, preventing them from exploiting growth opportunities available to comparable industries in the global market. Consequently, we expect growth opportunities to more strongly predict future growth in more financially open markets. Capital account openness The first panel in Table 6 presents estimates of the interaction between general capital account openness and exogenous growth opportunities in predicting future growth. The relation between growth and capital account liberalization is itself controversial. Rodrik (1998), Edison et al. (2002) claim that there is no correlation between capital account liberalization and growth prospects, whereas Edwards (2001), Bekaert, Harvey, and Lundblad (2005), and Quinn and Toyoda (2001) document a positive relation. Arteta, Eichengreen and Wyplosz 16

19 (2003) conduct robustness experiments using different measures of openness and conclude that the relation between growth and capital account openness is fragile. We focus on our largest sample to maximize the cross-sectional variation in our openness measures. Our measures of capital account openness are based on the IMF s Annual Report on Exchange Arrangements and Exchange Restrictions (AREAER). The first is an indicator variable that takes on a value of zero if the country has at least one restriction in the restrictions on payments for the capital account transactions category. The second measure, developed by Quinn (1997) and Quinn and Toyoda (2001), attempts to measure the degree of capital account openness; the measure is scored from 0 to 4, in half integer units, with a 4 representing a fully open economy. We transform Quinn s measure into a 0 to 1 scale. The measure is available for 48 of the 50 countries in our broadest sample. For both the IMF and Quinn measures of capital account openness, we find that the coefficient on the interaction between GGO MA and the associated capital account openness indicator is positive in all cases. However, the interaction coefficient is never statistically significant at the 5% level. Equity market liberalization In Panel B of Table 6, we explore the interaction effect between the exogenous growth opportunities measure, GGO M A, and indicators of equity market liberalization. Our first measure, the official equity market liberalization indicator, is based on Bekaert and Harvey s (2004) detailed chronology of important financial, economic, and political events in many developing countries. The variable takes the value of one when it is possible for foreign portfolio investors to own the equity of a particular country and zero otherwise. Industrialized countries, such as the United States, are assumed to be fully liberalized throughout our sample. Our second measure uses data on foreign ownership restrictions to measure the degree of equity market openness. Following Bekaert (1995) and Edison and Warnock (2003), the measure is based upon the ratio of the market capitalization of the IFC investable to the IFC global indices in each country. The IFC s global stock index seeks to represent the local stock market whereas the investable index corrects the market capital- 17

20 ization for foreign ownership restrictions. Hence, a ratio of one means that all of the stocks in the local market are available to foreigners. Accordingly, α i,1,t is a linear function of either the 0/1 indicator associated with official equity market liberalization or the continuous measure on the [0,1] interval capturing the degree of equity market openness. In contrast to the evidence for general capital account liberalization presented above, the link between growth opportunities and future output and investment growth is much stronger in economies that permit greater access to their equity markets. 10 The interaction coefficient is always statistically significant, both for the official equity market liberalization indicator and the liberalization intensity. The constant term is still positive, but no longer significant. This evidence suggests that there is a strong association between the ability to exploit global growth opportunities and the degree of foreign investor access to the domestic equity market. We also use the degree of stock market internationalization variable created by Levine and Schmukler (2003) as an indicator of equity market openness. effects are again positive, they are not statistically significant (not reported). 11 Banking openness While the interaction Finally, in Panel C of Table 6, we introduce a 0/1 indicator variable that captures periods when foreign banks are allowed to enter the domestic market either through the establishment of branches or subsidiaries or through the acquisition of local banks. Using a variety of sources, we have been able to determine important regulatory changes affecting foreign banks in 41 of our 50 countries over the past 23 years. The regression involving this new indicator therefore reflects a slightly smaller sample. The foreign banking liberalization indicator is equal to zero unless foreign have access to the domestic banking market through the establishment of branches or subsidiaries or through the acquisition of local banks (for details see Table 2 and Appendix Table A2). While recent studies have explored the impact of foreign banks on the efficiency and stability of the local banking sector (e.g. Claessens, 10 Henry (2000) and Bekaert and Harvey (2000) show that the growth rate in the capital stock increases on average following equity market liberalizations. 11 Note that the sample here started in

21 Demirgüç-Kunt, and Huizinga (2001)), our indicator variable is related to the regulatory environment foreign banks face with respect to establishing or expanding their operations in a local market. We also construct a first sign indicator that changes from zero to one when a country takes substantial first steps to improve access for foreign banks. Appendix table A2 lists the year of the banking liberalization for each of 41 countries. Similar to the equity market liberalization effect, there is a strong association between the opening of the banking sector to foreign banks and the ability to exploit exogenous growth opportunities. The interaction coefficients between both of the banking liberalization indicators and growth opportunities are always positive and statistically significant. 4 Capital Allocation and Growth Opportunities Apart from capital controls, there are many other country characteristics that may effectively segment markets, or prevent aligning growth opportunities with actual growth. In fact, until recently the growth literature seems to have largely ignored the potentially important role of financial openness. Instead, a vast literature documents a significant relationship between domestic banking development (e.g. King and Levine (1993)) or stock market development (e.g. Atje and Jovanovic (1989)) and economic growth. As Fisman and Love (2004) point out, the most obvious channel through which financial development would promote growth is through its role in allocating resources to its most productive uses. In the language of this paper, financial development helps align growth opportunities with growth. However, an influential paper of Rajan and Zingales (1998) has instead stressed the importance of financing constraints as the mechanism through which financial development promotes growth: industries heavily dependent on external finance grow faster in more financially developed countries. Interestingly, both articles assume a form of market segmentation to allow domestic financial development to play an important role in the inter-sectoral allocation of resources. As Bekaert, Harvey, and Lundblad (2005) point out, financial openness promotes financial development, so that this assumption may effectively ignore an important channel for allocative efficiency. In Section 4.1, we use our empirical framework to revisit this debate. 19

22 La Porta et al. (1997) have stressed the importance of investor protection and, more generally, the quality of institutions and the legal environment as sources for cross-country differences in financial development. In Section 4.2, we use our panel set-up to directly test the importance of investor protection in helping align growth opportunities with actual growth. We show that investor protection per se is less important than more general measures of political risk, specifically components of political risk which may be of particular importance for foreign direct investment. 4.1 Financial Development, External Finance Dependence, and Growth Panel A of Table 7 considers interaction effects with three important measures of domestic financial development: the ratio of private credit to GDP (banking development), equity market turnover and the ratio of equity market capitalization to GDP (both measures of equity market development). The coefficient on the interaction with the private credit ratio enters positively for both output and investment growth, and is significant at the 10% and 5% levels for each, respectively. However, the coefficients on turnover and size are actually negative in three of the four cases presented, but statistically insignificant for both output and investment growth in all cases. Together, this evidence suggests that domestic banking development is important for exploiting growth opportunities, but stock market development is not. This stands in contrast to the evidence presented above on stock market openness. Interestingly, this finding directly confirms the results in Fisman and Love (2004b). They postulate that the relation between actual growth in an industry in a particular country and its growth opportunities should be stronger depending on the level of financial development in the country. They test this hypothesis without measuring growth opportunities by investigating the correlation of industry growth rates across countries. They find that countries have correlated intersectoral growth rates only if both countries have high private bank credit to GDP ratios. Other measures of financial development do not yield significant results. The Fisman-Love test assumes the existence of globally correlated shocks, but ignores the presence of international capital flows. It is conceivable that international flows are the mechanism behind the correlation in cross-country sectoral growth rates not that these 20

23 countries simply have well functioning financial markets. Panel C (left side) provides some exploratory analysis of this issue. We split up our observations into four groups. First, we sort observations into below or above median financial development, using the private credit to GDP ratio, then into financially open and closed using the official equity market openness indicator. We regress GDP and investment growth on our measure of growth opportunities interacted with an indicator variable for each of the four groups. The results strongly support the idea that it is openness that drives the alignment of growth opportunities with growth, not financial development. Even in market with poor financial development, the interaction coefficient is highly significant as long as the country has a liberalized equity market. The GDP growth interaction coefficients are at least twice as large for open versus closed equity markets. Not surprisingly, a Wald test strongly rejects the equality of the open versus closed coefficients, while it fails to reject the equality of the coefficients for low versus high financial development. The Fisman-Love article casts doubt on the results by Rajan and Zingales (1998), who stressed the role of external finance dependence. We obtained industry-specific time-invariant measures of external finance dependence (the amount of investments not financed internally) and investment intensity (the ratio of investments to property, plant, and equipment) from Rajan and Zingales. These variables are based on US data and available only for manufacturing industries (see Rajan and Zingales (1998) for details). Using time-varying industry weights measured as an industry s relative value added in a given country, we construct aggregate measures of external finance dependence and investment intensity. Panel B in Table 7 provides a simple interaction analysis of the growth opportunities measure with these country-specific Rajan-Zingales measure. The interaction is positive but not overall statistically significant. This interaction effect appears inconsistent with the Rajan-Zingales hypothesis, as it implies that countries with a higher weight in industries that are heavily dependent on external finance manage to better align growth opportunities with growth. However, it is conceivable that industries which require much external finance are better represented in countries with well developed financial markets. This is exactly the claim made and supported by Fisman and Love (2004a). Apart from the debate between 21